The response of forest carbon storage to climate change is uncertain, contributing substantially to the divergence among global climate model projections. Numerous studies have documented responses of forest ecosystems to climate change and variability (e.g., annual or multi-annual droughts). However, the sensitivity of forests to climate variability -- in terms of both biomass carbon storage and functional components of tree species composition -- has yet to be quantified across a large region using systematically sampled data. We combined systematic forest inventories across the eastern USA, a species-level drought-tolerance index, and the Palmer Drought Severity Index (PDSI) to quantify relationships among forest biomass, community-mean-drought-tolerance, and PDSI in one-degree grid cells from the 1980s to the 2000s. To gain qualitative insights into the forest dynamics revealed by the empirical analysis of inventory data, we also performed complementary experiments with LM3-PPA, a trait-based dynamic global vegetation model. LM3-PPA simulates the demography of tree cohorts competing for light and water, and is designed to be integrated with GFDL Earth System Models. The LM3-PPA experiments were designed to understand how tree carbon allocation to wood, leaves, and fine roots is expected to shift across climate gradients and in response to episodic drought.
Analysis of inventory data showed that community drought tolerance increased and forest biomass decreased in grid cells where PDSI (water availability) decreased, and vice versa (decreased drought tolerance and increased biomass where PDSI increased). The response of forest biomass to PDSI was amplified by shifts in community-mean-drought-tolerance, because water stress tends to induce a shift in tree species composition towards more drought-tolerant but lower-biomass species. Thus, the observed response of forest biomass to PDSI was steeper than it would have been if species composition were constant over time. These results were statistically robust; i.e., accounting for potentially confounding variables (stand age, shifts in community-mean-shade-tolerance, and changes in selective harvest regimes) and alternative statistical modeling approaches led to similar conclusions. LM3-PPA simulations were broadly consistent with the analysis of forest inventory and suggest that species differences in carbon allocation to wood, leaves, and fine roots may contribute to the observed decrease in biomass with increasing community-drought-tolerance. Specifically, in LM3-PPA, competition drives plants to over-invest in fine roots when water is limiting, which amplifies the direct effects of drought on wood biomass. This result is corroborated by analyses of simpler competition models, as well as empirical experiments.